摘要
强化函数的设计是构建多智能体学习系统的一个难点。提出了一种基于知识的强化函数设计方法,根据实际应用的特点,将经验信息和先验知识引入到强化函数中,提高了强化学习的性能。通过在RobotSoccer中的应用和实验,基于知识的强化函数的学习效果要优于传统的强化函数。
The design of reward function is one of difficulties in building reinforcement learning system.A design of reward function based on knowledge is presented.In term of the characteristic of application system,the experience and the domain knowledge are introduced to reward function.The performance of reinforcement learning is improved.With the application and experiment of Robot Soccer,it is illuminated that the reward function based on knowledge has the better effect than the traditional reward function.
出处
《计算机工程与应用》
CSCD
北大核心
2005年第3期77-79,共3页
Computer Engineering and Applications
